Artificial intelligence (AI) services are now integral to our daily lives, influencing aspects such as job searches, housing, and relationships through AI-powered platforms. Many of these services employ federated learning (FL) systems to create personalized machine learning (ML) models for users, providing tailored predictions on interests like job offers, dating, and music videos. Despite the usefulness of FL systems, there is increasing evidence for their potentially harmful effects, such as boosting addictive user behavior or even genocide. This project breaks ground for trustworthy FL, shifting the focus of current FL research towards a more human-centric perspective. Besides the computational and statistical properties of FL systems, this project emphasizes important design criteria for trustworthy AI.

2024

2028